neuron-request@HPLABS.HP.COM (Neuron-Digest Moderator Peter Marvit) (10/10/88)
Neuron Digest Sunday, 9 Oct 1988 Volume 4 : Issue 12 Today's Topics: Re: Excaliber's Savvy What is MX-1/16? Pulse coded neural networks PDP under Turbo C Commercial Uses of Neural Nets: Survey Summary Washington Neural Network Society Meeting Announcement Hector Sussmann to speak on formal analysis of Boltzmann Machine Josh Alspector to speak on a neuromorphic learning chip Send submissions, questions, address maintenance and requests for old issues to "neuron-request@hplabs.hp.com" or "{any backbone,uunet}!hplabs!neuron-request" ------------------------------------------------------------ Subject: Re: Excaliber's Savvy From: bstev@pnet12.cts.com (Barry Stevens) Organization: People-Net [pnet12], Del Mar, Ca. Date: 01 Oct 88 14:56:24 +0000 With reference to Savvy: Die-hard neural network purists say that Savvy has nothing to do with neural nets. The explanation I got was that the system has no "connection" to recognized neural networks, such as back propogation. I asked the original author of the program what was under the covers. He wouldn't answer me directly, but suggested that I look up papers on something called "n-tuple pattern recognition", and that they were, among other places, found in the Sandia Labs library. I called Sandia, and since I'm not an employee, couldn't get a copy. I'd like to do some reading on the topic -- does anyone know where I can find papers on the topic : "N-tuple pattern recognition?" I have used Savvy. It has a very friendly front end, being as smart about English as you are with the synonyms you can build, it has an extremely powerful parsing capability that has let me experiment with extracting production rules from text, and performs nearest-neighbor classification on the words you want to look up very quickly. Whether or not this qualifies as neural network capabilities... UUCP: {crash ncr-sd}!pnet12!bstev ARPA: crash!pnet12!bstev@nosc.mil INET: bstev@pnet12.cts.com ------------------------------ Subject: What is MX-1/16? From: ghosh@ut-emx.UUCP (Joydeep Ghosh) Organization: UTexas Computation Center, Austin, Texas Date: 03 Oct 88 16:39:35 +0000 The DARPA Executive Summary on Neural Networks mentions a neural network simulation system called MX-1/16 with a projected storage capacity of 50M interconnects and processing speed of 120M interconnects/sec. Could someone shed light on who is building this machine, what network models does it support, system architecture, stage of development...? Thanks, Joydeep Ghosh Internet: ghosh@ece.utexas.edu University of Texas, Austin (512)-471-8980 [[ Maybe this is the fabled bee's brain (*smile*). Note for Joydeep's return address may need to be ghosh%ece.utexas.edu@cs.utexas.edu due to U. Texas internal rerouting. -PM]] ------------------------------ Subject: Pulse coded neural networks From: gill@nc.MIT.EDU (Gill Pratt) Date: Mon, 03 Oct 88 13:21:39 -0500 Our lab is still pursuing research on time interval coded nets, and we would enjoy corresponding with others doing the same. Gill Pratt gill@mc.lcs.mit.edu ------------------------------ Subject: PDP under Turbo C From: uunet!otago.ac.nz!R_OSHEA Date: 04 Oct 88 09:50:13 -0800 I am trying to recompile PDP models under Turbo C Version 1.5. Currently I have not got a curses.h header file for Turbo C and do not know Turbo C graphics primitives well enough to translate the curses header file I do have into Turbo. Could anyone be of any assistance here. Yours Aaron.V.Cleavin University Of Otago P.O. Box 56 New Zealand. Replies should go via Robert O'Shea uucp: r_oshea%otago.ac.nz@uunet.uu.net or ...!uunet!otago.ac.nz!r_oshea internet: r_oshea@otago.ac.nz csnet: R_OSHEA%OTAGO.AC.NZ@RELAY.CS.NET or R_OSHEA%OTAGO.AC.NZ%waikato.ac.nz@RELAY.CS.NET [[ I assume Aaron means Rumelhart & McClelland's PDP Vol3, "Explorations". I remember a request about Macintosh versions of this simulator software. Any help on either count from readers? -PM]] ------------------------------ Subject: Commercial Uses of Neural Nets: Survey Summary From: bstev@pnet12.cts.com (Barry Stevens) Organization: People-Net [pnet12], Del Mar, Ca. Date: 04 Oct 88 22:36:13 +0000 Two weeks ago, I posted a message that I had done a survey of companies looking for applications that were suitable for neural networks, and asking if there was any interest. Since that time, the responses have been coming back by Email almost daily, just over 30 of them so far. Accordingly, I have prepared a summary of that study for posting in comp.neural-nets. The summary appears as a comment to this message, and is approximately 200 lines in length. The original report can't be released, since it contains some proprietary material. The summary, however, contains material that is available from numerous public sources, if one knew where and when to look. To keep the post down, I had to exclude all of the technical detail that usually accompanies discussion about neural networks. I describe the applications that were identified, and what, if anything, has been done about them. Period. Barry Stevens UUCP: {crash ncr-sd}!pnet12!bstev ARPA: crash!pnet12!bstev@nosc.mil INET: bstev@pnet12.cts.com [[ I've talked to Barry and his background is the proverbial engineer's. Several issues ago, someone asked what Neural Nets are "good for" and how they are really being used. Barry found quite a bit. Due to mailer problems, he couldn't send me his survey directly; I'm taking it from his USENET posting and it will appear next issue. -PM]] ------------------------------ [[ Editor's Note: I will try to put conference and talk announcements at the end of the Digest so that folks can parse the conents of the Digest more easily. Let me know what you think. -PM ]] Subject: Washington Neural Network Society Meeting Announcement From: weidlich@ludwig.scc.com (Bob Weidlich) Date: Thu, 29 Sep 88 23:18:58 -0400 The Washington Neural Network Society First General Meeting October 12, 1988 7:00 PM Speaker: Fred Weingard Booz, Allen & Hamilton, Inc. Arlington, Virginia. Neural Networks: Overview and Applications Neural networks and neurocomputing provide a novel and promis- ing alternative to conventional computing and artificial in- telligence. Conventional computing is characterized by the use of algorithms to solve well-understood problems. Artifi- cial intelligence approaches are generally characterized by the use of heuristics to obtain good, but not necessarily best, solutions to problems whose solution steps are not so well-understood. In both approaches, knowledge representions or data structures to solve the problem must be worked out in advance and a problem domain expert is essential. These ap- proaches result in systems that are brittle to unexpected in- puts, cannot adapt to a changing environment, and cannot easi- ly take advantage of parallel hardware architectures. Neural network systems, in contrast, can learn to solve a problem by exposure to examples, are naturally parallel, and are ``robust" to novelty. In this talk Fred Weingard will give a general overview of neural networks that covers many of the most promising neural network models, and discuss the applica- tion of such models to three difficult real-world problems -- radar signal processing, optimal decisionmaking, and speech recognition. Fred Weingard heads the Neural Network Design and Applications Group at Booz, Allen & Hamilton. Prior to joining Booz, Al- len, Mr. Weingard was a senior intelligence analyst at the De- fense Intelligence Agency. He has degrees in engineering from Cornell University and is completing his doctorate in computer science / artificial intelligence at George Washington Univer- sity. The meeting will be held in the Contel Plaza Building Audito- rium at Contel Federal Systems in Fairfax, Virginia, at the southwest edge of the Fair Oaks mall. Directions from 495 Beltway: Take Route 66 Westbound (toward Front Royal) and get off at route 50 heading west (Exit 15 Dulles/Winchester). Go 1/4 mile on route 50, follow sign to "shopping center". Stay in right lane and merge into service road that circles shop- ping center. Take driveway from service road to Contel build- ing. Address is 12015 Lee Jackson Highway. Contel building is across shopping parking lot from Lord and Taylor, near Sears. For further information call Billie Stelzner at (703) 359-7685. Host for the meeting is the recently-established Contel Technology Center. Dr. Alan Salisbury, Director of the Technology Center, will present a brief introduction to the plans for research and application of technology at the Contel laboratory, including work in artificial intelligence and man-machine interface design. Schedule: 7:00 - 7:15 Welcoming (Alan Salisbury) 7:15 - 8:15 Speaker (Fred Weingard) 8:15 - 8:30 Report on Neural Network Society (Craig Will) 8:30 - 9:30 Reception, informal discussion ------------------------------ Subject: Hector Sussmann to speak on formal analysis of Boltzmann Machine From: pratt@zztop.rutgers.edu (Lorien Y. Pratt) Organization: Rutgers Univ., New Brunswick, N.J. Date: Fri, 30 Sep 88 20:51:11 +0000 Fall, 1988 Neural Networks Colloquium Series at Rutgers On the theory of Boltzmann Machine Learning ------------------------------------------- Hector Sussmann Rutgers University Mathematics Department Room 705 Hill center, Busch Campus Friday October 14, 1988 at 11:10 am Refreshments served before the talk Abstract The Boltzmann machine is an algorithm for learning in neural networks, involving alternation between a ``learning'' and ``hallucinating'' phase. In this talk, I will present a Boltzmann machine algorithm for which it can be proven that, for suitable choices of the parameters, the weights converge so that the Boltzmann machine correctly classifies all training data. This is because the evolution of the weights follow very closely, with very high probability, an integral trajectory of the gradient of the likelihood function whose global maxima are exactly the desired weight patterns. - -- - ------------------------------------------------------------------- Lorien Y. Pratt Computer Science Department pratt@paul.rutgers.edu Rutgers University Busch Campus (201) 932-4634 Piscataway, NJ 08854 ------------------------------ Subject: Josh Alspector to speak on a neuromorphic learning chip From: pratt@zztop.rutgers.edu (Lorien Y. Pratt) Organization: Rutgers Univ., New Brunswick, N.J. Date: 04 Oct 88 17:58:21 +0000 Fall, 1988 Neural Networks Colloquium Series at Rutgers Electronic Models of Neuromorphic Networks ------------------------------------------ Joshua Alspector Bellcore, Morristown, NJ 07960 Room 705 Hill center, Busch Campus Piscataway, NJ Friday October 21, 1988 at 11:10 am Refreshments served before the talk Abstract We describe how current models of computation in the brain can be physically implemented using VLSI technology. This includes modeling of sensory processes, memory, and learning. We have fabricated a test chip in 2 micron CMOS that can perform supervised learning in a manner similar to the Boltzmann machine. The chip learns to solve the XOR problem in a few milliseconds. Patterns can be presented to it at 100,000 per second. We also have demonstrated the capability to do unsupervised competitive learning as well as supervised learning. - -- - ------------------------------------------------------------------- Lorien Y. Pratt Computer Science Department pratt@paul.rutgers.edu Rutgers University Busch Campus (201) 932-4634 Piscataway, NJ 08854 ------------------------------ End of Neurons Digest *********************